Segmentation of Retinal Blood Vessels through Gabor Features and ANFIS Classifier

We introduce a technique for extracting the vessel structure in the fundus image of a retina. Retinal vessel segmentation achieved by categorizing every pixel belonging to vessel structure or not, derived from characteristic vector consisting of the gray level values and coefficients of 2-D Gabor wavelet at various scales. Specific frequency tuning of Gabor wavelet allows vessel segmentation even in the presence of noise in the image. We use Adaptive Network Fuzzy-Inference System (ANFIS) classifier with linguistic expression modeling capability and self-learning, yielding accurate classification. The openly accessible DRIVE database is used for performance evaluation of manually labeled images. In addition, Performance evaluation carried on fundus images provided by the ophthalmologist. The results obtained are quiet promising when inspected visually in both normal as well as pathological images.